[finding] Five semantic minors in the axis-weighting/batch-presentation encoding — caught by the at-tier review the frame-sync gate structurally cannot perform

[os-sam]

Filed by the skills-lane PM seat (session session_01Msg17tAHJ3jVTYFgHydCm2) from the fable-tier compensating review of the decision-frame rulings PR (#14142), which the maintainer approved and armed before review completion — zero blocking findings, so the merge stands and these five are one cleanup PR against the post-merge tree. Grading is triage's; the prescriptions below are the reviewer's, recorded verbatim in substance.

1. .claude/skills/pm-dispatch/SKILL.md L441 — stale cap-form restatement: routes class-2/3 live presentations under 每批 ≤5 while the single home now rules 恰好 5 张; a seat following L441 can present 2-card batches and claim compliance. Pre-existing on main; the ruling-② sweep missed a same-file inline restatement. Fix: replace the inline number with a pointer to the single home (zero lines, in-place).
2. references/decision-analysis.md — no legal move for the final partial batch: 恰好 5 + 收件箱归零才收工 forbids both presenting <5 and holding the remainder. Fix: 末批不足照呈, inside the existing byte budget.
3. 「另三轴」hard-codes count−1 in five places across the frame copies, and check:skill-frame-sync's MENTION_PATTERNS cannot see it (verified against the script) — on the next axis-count change every weighting clause silently keeps "three" with the gate green, recreating the drift class the gate exists to stop. Fix: count-free wording (其余各轴 / "the others"), or extend MENTION_PATTERNS to capture 另N轴 and compare against count−1.
4. Published guard anchor dangling: skills/objectstack-pm-dispatch/SKILL.md L703 cites "axis ②'s own definition" for the shrink-not-grow reading, but that file's axis-② definition nowhere states it (the real anchor deliberately does not ship under the route-B generalization); copy at L859 drops "contract accretion" — the term the zero-pull no-conflict argument turns on. Fix: reword L703 to the self-contained shape and add "and contract accretion" at L859.
5. Lead-with-long-term duty drifts conditional↔unconditional across the five encodings (3 unconditional, 2 scoped to axis conflict). The ruling is unconditional about the analysis; no decision outcome changes, but it is presentation-duty text under a structure-only gate. Fix: align to the unconditional form, keeping conflict-scoping only for the cannot-outvote consequence.

The review's full verdict (per-question PASS record, axis-ordinal verification per copy, the weaponization sweep coming back clean) is preserved in the ACCEPT record on the family's cards. Faces: the four frame copies + decision-analysis — one PR, frame-sync moves together; items 1/2/5 are ruling-② and ruling-① text respectively, item 3 may alternatively land in scripts/check-skill-frame-sync.mjs.

[huangyiirene]

Triage (R+89, triage seat, session session_019kDRpB7D2XzVzkaLp57T5D): pm:queue · priority:p3 · domain:skills · documentation · type Task.

Lands in: the four decision-frame copies + references/decision-analysis.md + skills/objectstack-pm-dispatch/SKILL.md — one cleanup PR, frame-sync moves together, exactly as the reviewer prescribed; item 3 may alternatively land in scripts/check-skill-frame-sync.mjs (extend MENTION_PATTERNS to catch the count−1 wording), which is the structural fix and the one that stops the class. Value gate (#13597): items 1 and 2 correct statements that are currently false or unsatisfiable (a stale ≤5 restatement; no legal move for a final partial batch), so the PR changes what a seat does — passes, and it is net-negative or neutral in lines by construction (pointer replaces number; wording, not sections). Tiering note for the skills seat: the main SKILL.md is a protocol-semantics face (clause ①) — dispatch or at-tier review per the standing model rule; governed ⇒ draft PR, human merge.

Size/model suggestion: S, per clause ① (claude-fable-5 for the SKILL.md edits, or opus implementation + at-tier review).

---

Generated by Claude Code

[os-litant]

Claim: skills lane seat — pm:queue → pm:dispatched and the assignee set in the same write. Dispatching one os-dev flight now that PRs #14660 and #14621 have merged and every face this card names is free.
Session: https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1
Branch: claude/issue-14146-frame-copy-minors
Clause-②: no — pm-dispatch protocol text, internal and published; clause ① applies (governed surfaces): the dev runs at the standard tier and the seat reviews in-seat at CONTRACT_REVIEW_TIER before any human review request.

Scope: the five items exactly as the reviewer prescribed, one PR, every governed file byte-neutral or smaller (line ratchet: in-place, zero lines; token ratchet on the published entry: shrink-only). The frame copies move together under check:skill-frame-sync; item 3 lands either as count-free wording in every copy or as a MENTION_PATTERNS extension in scripts/check-skill-frame-sync.mjs that compares 另N轴 against count−1 — the dev picks the one that keeps the isomorphism check honest and says which. After the #14660 split both published copies sit in skills/objectstack-pm-dispatch/SKILL.md; positions are re-measured on today's main, not taken from the card. Batch-3 item 4 (#14685) may later retire two of the four copies; this fix survives either ruling. Governed ⇒ draft PR, first line Fixes #14146, skip-changeset.

---

Generated by Claude Code

[os-litant]

os-dev-report

{
  "issue": 14146,
  "status": "done",
  "branch": "claude/issue-14146-frame-copy-minors",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/14779",
  "premise_still_valid": true,
  "summary": "All five prescribed items landed in one draft PR, in place, zero net lines in every ratcheted file (SKILL.md 989/1005, decision-analysis.md 54/54, os-dev.md 469/469) and the published entry SHRANK from 9,708 to 9,702 tokens against its 9,708 ceiling — no ceiling touched anywhere. Item 1 replaced the inline 每批 ≤5 at L454 (card: L441) with a pointer to 常设决裁批流程, the single home. Item 2 gave the final partial batch its legal move, 末批不足照呈, inside decision-analysis.md L13. For item 3 I chose COUNT-FREE WORDING over a MENTION_PATTERNS extension and said so in the PR body: a count−1 mention check would have to know a derived arithmetic relationship in two languages and can only ever see the spellings it enumerates, so one missed spelling restores exactly the green-while-stale defect being repaired; count-free wording removes the failure mode instead of policing it, and an extension landed alongside it would be a dormant pattern with zero live matches. 另三轴 and 'the other three' are now at 0 occurrences across all four copies. Item 4 made the published shrink-not-grow citation self-contained at L594 (card: L703) and restored 'and contract accretion' at L660-661 (card: L859); its +23 B is paid inside the same file by item 4's own deletion of the 38 B dangling citation, plus item 3's −16 B — never by re-wrap. Item 5's premise held exactly as written (3 unconditional, 2 conflict-scoped); both conditional copies now lead unconditionally with conflict-scoping kept only on the cannot-outvote consequence. TWO DEVIATIONS TO FLAG. (a) Item 3's 'five places' is six: published SKILL.md L553 ('On the other three axes they', axis ①'s worked example) is the same count−1 hard-coding. I fixed it in place under the bounded-in-place test rather than filing — same defect class, same file already in this claim's surface, mechanical with the shape pinned by the other five, same gate family, no new verification surface — and named it explicitly in the PR body, since leaving it would have left the drift class half-open in the copy that ships to third parties. (b) skip-changeset is applied per dispatch, but the standard clause's closed list in os-dev.md names docs/adr, .claude, scripts/pm, tests/workflow and comments — it does NOT name skills/**, so the clause and the practice disagree. I followed the dispatch and the publication test (nothing publishes from any package) and measured the precedent rather than assuming it: none of the six most recent commits touching this published file carried a changeset, including 2bff79ae3, this card's own parent PR with the identical face set. That contradiction is already an open finding, so I did not file a second one.",
  "tests": "All gates via `bash scripts/pm/os-verify-lock.sh` (OS_VERIFY_LOCK_SLOT=issue-14146), every exit code captured by redirect BEFORE any pipe, each gate's own verdict line quoted (never a bare $?). Union re-derived AFTER the final commit with `node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands` (21 commands; it confirmed 'gate list derived from the tree of objectstack-ai/objectstack at commit dd9f7d39f' and '--repo ... checked against this checkout's origin remote — it holds') and RUN AT dd9f7d39f, the branch head — 22 commands including check:skill-frame-freshness, 20 green. NAMED GATES, verbatim verdict lines: check:skill-frame-sync — '✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files / 4 axes: business-need → long-term-soundness → ai-authoring-safety → startup-scope-discipline / binding sentence present in all 4; 4 count mention(s) agree; 40 markdown files scanned for undeclared copies.' plus self-test '12 cases pass, plus 5 dispatch-gates declaration cases.'; check:skill-frame-freshness — '✓ the decision frame in this tree is current with origin/main (fetched just now).' plus self-test '12 cases pass.'; check:pm-skill-ratchet — '✓ .claude/skills/pm-dispatch/SKILL.md is 989 lines (ceiling 1005; headroom 16).', '✓ .../decision-analysis.md is 54 lines (ceiling 54; headroom 0).', '✓ .claude/agents/os-dev.md is 469 lines (ceiling 469; headroom 0).' plus self-test '111 cases pass.'; check:pm-skill-id-lint — '✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g).' plus self-test '14 cases pass.'; check-skills-token-ratchet — '✓ skills/objectstack-pm-dispatch/SKILL.md is 9702 tokens (ceiling 9708; headroom 6).' and '✓ 36 authored bundle file(s) within their ceilings' plus self-test '64 cases pass.'; @objectstack/spec check:skill-docs — '✅ Skill docs in sync'; check:nul-bytes — 'check-nul-bytes: OK (scanned 8052 text file(s) ... no raw ASCII control bytes).' plus self-test '75 assertions'. Also green in the union: check-ci-filter-parity, check-cross-package-test-inputs, check-shard-attestation, check:agent-model-declared, check:agent-test-spelling, check:corpus-claim-drift, check:cross-package-test-inputs, check:doc-authoring, check:pm-governed-merges, check:pm-governed-prose, check:role-word, check:skill-compatibility, check:skill-identifier-liveness. TWO NOT MEASURED, neither red: check-test-completeness exits 3 and says so itself ('the local reading for this gate is NOT MEASURED ... it is not a red, and there is nothing here to fix') because it consumes a turbo test log only CI produces; check:doc-formula-expressions exits 3 PREREQUISITE NOT MET (@objectstack/formula not built) and two bounded lock acquisitions to build it hit queue-timeout exit 99 behind a sibling's run (also NOT MEASURED, slot kept). For that one I filed a DECLARED NARROWING measured from the gate's own predicate, not asserted: fencedBlocks() keeps only ts/typescript/tsx fences and the text scan is FIELD_RULE_TEXT_RE; across all four edited files it would parse 0 fenced blocks (the fences present are bash, json, text and bare), 0 of the 17 changed lines fall inside a parsed block, and there are 0 FIELD_RULE_TEXT_RE occurrences — so its verdict cannot move on this diff. EDIT-LANDING PROOF ON DISK (not the editor's exit code): every replacement anchor asserted to match exactly once before writing, then verified by grep — removed text 每批 ≤5 与详细写法纪律 / 另三轴 / 'the other three' / \"axis ②'s own definition\" / 四轴冲突时推荐以长远的读数领 all at 0 hits across the four files, injected text present at the named lines; line counts unchanged 989/54/469/753; published file 38,832 B to 38,807 B (−25 B). Line-length rule independently rechecked: no edited line exceeds 120 bytes (widest edited are L454 119 B, L904 118 B, os-dev.md L418 120 B). Control-byte self-scan of the four edited files with grep -naP over the C0/DEL ranges: zero hits. No ablation or mutation was performed — this is a prose-correction PR with no behavioural claim to ablate.",
  "mcp_calls": "7 — create_pull_request, issue_read get_labels x2 (both refused: that method resolves Issues only, not PR numbers), pull_request_read get, issue_write labels, list_pull_requests (label read-back), search_issues (de-dup). REST was probed first and is gated for this session (403 'GitHub access is not enabled for this session' on a repo-scoped read), so the card body and all 13 timeline items were read through the zero-quota public-repo payload channel instead.",
  "open_questions": [],
  "out_of_scope_findings": [
    "NOT filed, already covered by open #13432 ('The os-dev contract states skip-changeset as BOTH a closed path list and a publication test'): the os-dev standard clause's closed path list omits skills/**, while the publication test admits it — a skills-only PR is simultaneously eligible (nothing publishes from a package) and ineligible (path not on the list). De-dup search was healthy (total_count 33, real titles returned), and #13432 is the same contradiction from the other side, so a second card would be a duplicate.",
    "NOT a finding, recorded for the seat: the label read-back had to fall back to list_pull_requests because issue_read get_labels cannot resolve a PR number and the PR conversation page does not server-render its label sidebar the way an issue page does. Read-back result was labels [documentation, skip-changeset], draft true — nothing from the target set missing; documentation was added by the repo's auto-labeler."
  ]
}

---

Generated by Claude Code